Self-driving cars have been a near-future dream for long enough to question the “near” part of that statement. However, like any advanced new technology under development, a variety of supporting technologies must be refined to enable the end-product. Besides raw computing power, there is perhaps nothing more critical to vehicles that can drive themselves than the ability to obtain reliable information from the outside world.
This means not just one sensor, or even one type of sensor, but a variety of systems to give an overall picture of the immediate surroundings. These may include:
• GPS receivers to pinpoint a car’s location on the ground
• Cameras to capture the immediate environment as our eyes do
• Ultrasonics for close-range perception
• Radar to detect far-off objects
But the sensor that has garnered more attention than any other for this developing industry is “lidar,” or light detection and ranging. Similar to how radar scans with radio-frequency waves, lidar uses focused visible light to scan the environment, creating a “point cloud” representation of 3D surroundings. With a plethora of lidar startups emerging, and different techniques being refined, what can we expect in the “near-future” for this sensing technique?
Lidar Prices Become More Affordable
Consider that a decade ago, a lidar unit could set you back $75,000, but by 2017, these devices had dropped to under $10,000. Today, multiple manufacturers are touting sensors with prices in the hundreds of dollars, such as Velodyne’s H800 unit that has a high-volume target of less than $500 each. Others are targeting even lower prices, with the idea that lidar manufacturing will follow a similar price curve as computer chips, which have become both exponentially more powerful and cheaper over the years.
Of course, such (relatively) inexpensive devices available today aren’t always an apples-to apples comparison to their ancestors. These newer hundreds-of-dollars models are generally limited to a scanning range in the order of 100º, not the full 360º rotation capability of more expensive devices. This, of course, gives automotive manufacturers options, such as lower-level autonomy with two limited-range sensors for front and back, or a full range of scanning with four.
While still a significant cost at $500 for one (up to $2000 for four), this isn’t the deal-breaker that $75,000 might be. Factor in potential repair costs and even if you’re willing to pay $100,000+ for a car, tens of thousands of dollars to repair a sensor is still quite a hit to the wallet.
Lidar Size Improvements: Smaller, Directional & More "Solid"
Besides lower price, another benefit to such directional lidar is that an automobile doesn’t need a “lidar probe orb thingy” sticking out of the roof. These new units can instead be made small enough to squeeze into the interior of the car, or in subtly protruding sections that could appear similar to a signal light.
Part of the reason that these sensors can be miniaturized is their use of solid-state electronics and/or directional mirrors that aim the lasers. Rather than spinning the entire laser assembly, such mirrors can aim light beams in a similar manner to projection television of old in both the horizontal, and–a generally more limited–vertical angle.
In the case of solid-state and MEMS-based (microelectromechanical systems) devices, light can be aimed like a phased array radar installation, changing direction using carefully timed pulses of light from multiple sources. While that’s a bit of a simplification of the process, and there are multiple techniques being explored, this general idea will likely be the gateway to truly inexpensive “lidar-on-chip” technologies.
Such solid state-techniques were explored at MIT in 2016, shrinking a lidar unit to a size dwarfed by a US dime. Now we’re seeing this sort of tech in actual products. This ideal may take the form of MEMS, controllable liquid-crystal metasurfaces (LCM), optical phased arrays (OPA), or perhaps even some combination. Kyber Photonics, an MIT spinoff founded in 2020, is even working on a steering technique inspired by the 1960s-era Rotman lens. So while the technology may be rapidly maturing, how it will ultimately pan out is still very much an open question.
Lidar Software Options
At its heart (shine a laser and time how long it takes to return, to calculate distance to target) lidar is a relatively straightforward measurement technique. A huge part of what will set one stem apart from others is the software running the show behind the scenes.
For example, lidar manufacturer Luminar’s Hydra system features ongoing software updates that promise to make the system more capable as time progresses. To make sense of the data it receives as a point cloud, it employs a powerful 8-Core ARM processor with a 512-core NVIDIA GPU.
AEye, another lidar-based company, uses inputs from a visual camera to help determine where the actual lidar measurement should focus. This system, which they call “iDAR” for intelligent detection and ranging, claims to collect more relevant data while reducing power consumption of the system. The system captures what they call “vixels,” a combination of “video” and “voxel,” (~volume pixel), for even more information about the environment.
However this technology shakes out, one thing is certain: It will be a whole new level of acronyms and portmanteaus to learn.
Lidar Range and Adoption
In automotive applications, we can expect lidar sensing ranges to keep expanding. Radar systems are seen by some (i.e. Elon Musk) as being the proper solution for long-range sensing at highway speeds. Current lidar systems, however, are often touted as having a range of 200-300 meters. Traveling at 70mph, a 200m sensing range gives an automobile well over six seconds to react to a situation.
In a controlled test, Aeye has demonstrated their lidar working at 1000+ meters, which would theoretically give an automobile over 30 seconds of reaction time. Luninar claims a 500m max range with 10% or greater reflectivity, which works out to over 15 seconds at 70mph. While both claims have their limitations, expect the usable, or “real-world” ranges to keep increasing. As this happens, expect radar to lose some of its allure in self-driving applications. Perhaps we’ll even see Tesla eyeing this tech in the future… though predicting their moves can often be an exercise in futility.
Of course, we shouldn’t be “laser focused” on automotive systems alone, as this tech has been miniaturized to the point where it’s now included on the iPad Pro as well as the iPhone 12 Pro. It’s likely that we’ll see this tech integrated even more into robotics and electronics applications for short- and medium-range measurement. Whether for advanced driver-assistance systems (ADAS) now, fully self-driving automobiles in the future, or any number of other applications, it will be exciting to see how this tech develops and evolves over time.
